Apple’s M2 Tested On a Linux System, Loses to its Intel and AMD Counterparts

Most of us have seen an Apple chip tested on the Mac OS platform. How about it being tested on the Linux OS for a much more fair comparison? Your wishes have been granted! Michael Larabel from phoronix tirelessly compiled various benchmarks featuring the M2 from Apple on Arch Asahi Linux. For comparison purposes, Intel and AMD were also brought in to equation.

This gives us a chance to view how well Apple’s Mac OS is optimized and what exactly gives it a performance bump other than its ARM-ISA. Can this test show the actual raw power of Apple’s silicon without any external interferences? In a way it can, although being optimized specfically for Macintosh OS, degradation in performance is expected.

The initial test bench consists of the following CPUs:

  • Apple’s M2 
  • Intel i7-1280P (Alder Lake)
  • AMD R9 5900HX (Rembrandt, Zen3+)

How do we get Apple to work on another platform?

 

Asahi Linux is a modified form of Linux allowing for Apple chips to be used on the Linux OS. The Intel and AMD laptops were tested on the Arch Linux OS, whereas Apple’s M2 used the Asahi Linux OS.

Apple’s M2 MacOS vs LinuxOS | Michael Larabel at phoronix

Now, the question arises “How does the Linux OS version fare against Apple’s own MacOS?”. Luckily for us, many benchmarks were compiled.

Apple MacOS vs Linux OS

As seen, both platforms trade blows with one another. Where various tests make use of a better compiler on the Linux OS and others favor Apple’s optimization. It is actually impressive to see such an early state of the Asahi Linux OS performing so well. 

WebP Image Encoding Performance (Lossless) | Michael Larabel at phoronix
 
simdjson 2.0 (Kostya) Performance | Michael Larabel at phoronix
Renaissance 0.14 (ALS Movie Lens) Performance | Michael Larabel at phoronix
x265 3.4 (Bosphorus 1080p) Performance | Michael Larabel at phoronix

Apple vs AMD vs Intel performance on the Linux platform

The various laptops used for this test are as follow:

  • Apple Mac Mini – M1 8GB 
  • Apple MacBook Air – M2 8GB 
  • MSI MS-14C6 Evo Notebook – Intel Core i7 1280P (Alder Lake)
  • ASUS G512QY – Ryzen 9 5900HX 
  • ThinkPad X13 Gen3 – Ryzen 7 PRO 6850U (Rembrandt, Zen3+)

Now bear in mind, Apple will always take the lead in efficiency due to the use of ARM. However, for these tests we do not have exact performance/watt metrics due to the lack of power consumption indicators. 

Starting off, we have Etcpak 1.0. Etcpak is basically a texture compression utility. It is considered as one of the fastest in its league. The benchmarks show a clear performance difference between the 3 giants. It may be due to the lack of necessary optimization for Apple. However, a performance difference of over 200% is not to be taken lightly.

Etcpak 1.0 (Multi-threaded Test) (ETC2) Performance | Michael Larabel at phoronix
Etcpak 1.0 (Single-threaded Test) (ETC2) Performance | Michael Larabel at phoronix

In fairness to Apple, many AI-based tools will almost always perform better on their platform due to Apple’s Neural Engine. Intel plans to develop something just like this with its 14th Gen (2023-24) Meteor Lake CPUs.

LeelaChessZero gave the lead back to Apple. LeelaChessZero is a deep neural network–based chess engine that offers near-human levels chess gameplay. Apple takes a massive lead of over 250% here, although the M2 chip falls behind by a huge margin (~50%).

LeelaChessZero 0.28 (Eigen) Performance | Michael Larabel at phoronix

Moving on to WebP Image Encode 1.1, we see Apple lose its lead once more by a margin of around 25%.

WebP Image Encode 1.1 (Lossless) Performance | Michael Larabel at phoronix

simdjson is a JSON parser (breaking down instructions) that runs at gigabyte/second speeds capable of parsing millions of JSON documents in 1 second. Here Intel and AMD usurp Apple’s throne yet again.

simdjson 2.0 (Top Tweet) Performance | Michael Larabel at phoronix

Xmrig is an open source crypto-currency miner that uses CPU horsepower to mine Monero. Those interested in CPU mining (GPU shortage PTSD flashbacks), will find Apple’s offerings to be relatively decent based on their low power consumption. However, the day when your profit surpasses your initial investment will be very far away.

Xmrig 6.12.1 (Monero-1M) Performance | Michael Larabel at phoronix

Java Gradle Build is an open source tool that is used to build/develop/test applications and libraries for various langauges. Intel falls behind heavily in this workload, losing out to Apple by a massive ~50%.

Java Gradle Build (Reactor) Performance | Michael Larabel at phoronix

Video Encoding Performance

VP9 is the codec utilized by Youtube allowing for the best looking image quality. In the VP9 Encoding test, the R9 5900HX pulls ahead of the pack by more than 2x. Intel once again, fails to deliver comparable performance. 

VP9 libvpx Encoding 1.10.0 (Boshphorus 4K) Perfomance | Michael Larabel at phoronix

Moving on to x265, Intel gains back its position. However, Apple’s M1 falls behind by almost 8x as compared to AMD’s R9 5900HX. 

x265 3.4 (Bosphorus 1080p) Performance | Michael Larabel at phoronix

ASKAP and Encryption Performance

ASKAP is used to benchmark the performance of a CPU in various algorithms to image data from the Australian SKA Pathfinder. Here Intel is seen at the number 1 spot, where AMD takes a massive blows. Apple’s M1 and M2 fall in the decent category.

ASKAP (tConsole – Gridding) Performance | Michael Larabel at phoronix

Cryptsetup is a utility used to conveniently set up disk encryption based on the DMCrypt kernel module. In this encryption test, Apple’s M1 and M2 lose their footing, though not by much. 

Cryptsetup (AES-XTS 256b Encryption) Performance | Michael Larabel at phoronix

The Conclusion

The full benchmark should be read at phoronix for a much more detailed overview. Although, to put it shortly Apple’s M2 is more often than not on par with its Intel and AMD counterparts.

  1. AMD R9 5900HX
  2. Core i7 1280P
  3. Apple M2
A Pie chart of winning percentages | Michael Larabel at phoronix

It is indeed true that Apple may have lost (not by much) in these Linux based benchmarks but if we had access to the efficiency numbers, Apple would’ve won in every front. Intel and AMD based on x86-64 will always fall behind the ARM based Apple. 

What’s surprising is that the Linux and Macintosh OS benchmarks are nearly neck to neck. Such support for a really sophisticated CPU is indeed breathtaking. Kudos to the developers! 

Abdullah Faisal
With a love for computers since the age of give, Abdullah has always sought to delve into the depths of information, and uses it as his guiding light. He believes success is of utmost importance as history is written by the victor.